DreamDIA-XMBD

DreamDIA-XMBD analyzes data-independent acquisition (DIA) proteomic datasets to improve peptide identification and quantification by extracting deep representation features from peptide elution patterns.

Key Features:

• DIA proteomic data processing: Supports analysis of data-independent acquisition (DIA) proteomic datasets for peptide identification and quantification.
• Deep representation network: Extracts deep representation features from elution patterns of target peptides and captures information across multiple ions.
• Expanded theoretical elution profiles: Extracts features from dozens of theoretical elution profiles per precursor ion compared with the 6–10 selected transitions typically used by OpenSWATH, Skyline, and DIA‑NN.
• Non-linear discriminative modeling: Processes extracted features using non-linear discriminative models.
• Positive–unlabeled learning with decoys: Employs a positive–unlabeled learning framework that incorporates decoy peptides as affirmative negative controls to distinguish true targets from false positives.
• Performance focus: Targets improved identification and quantification performance in DIA analyses.
• Implementation: Implemented in Python.

Scientific Applications:

• Peptide identification in DIA proteomics: Enhances detection of target peptides in DIA datasets.
• Peptide quantification in DIA proteomics: Improves quantitative accuracy for precursor and peptide-level measurements.
• Specificity and coverage improvement: Reduces false positives and increases coverage in proteomic analyses through expanded elution-profile features and decoy-based discrimination.
• Comparative benchmarking: Facilitates comparison of identification and quantification performance against tools such as OpenSWATH, Skyline, and DIA‑NN.

Methodology:

Extracts deep representation features from elution patterns across multiple ions and dozens of theoretical elution profiles per precursor using a deep representation network, then processes these features with non-linear discriminative models within a positive–unlabeled learning framework incorporating decoy peptides as negative controls.